Research on University Books Bibliography Recommendation System Based on Collaborative Filtering Algorithm
Abstract
In order to realize personalized recommendations in university library borrowing systems, this study constructs a multi-dimensional implicit interest model based on borrowing duration, total borrowing frequency, and renewal records. To address the severe data sparsity of the user-item borrowing matrix, a hybrid collaborative filtering (CF) algorithm integrating K-nearest neighbor (KNN) search and a Chinese Library Classification (CLC) tree-based type factor is proposed. Furthermore, a five-tier bibliographic recommendation system architecture is designed, implementing 12 functional modules across foreground and background partitions using the Hadoop distributed framework with an offline/online decoupled strategy. Extensive experiments on actual borrowing data from 1,000 students (540,000 records) demonstrate that the proposed hybrid model reduces the Mean Absolute Error (MAE) by 18.7% compared to traditional User-based CF and by 23.4% compared to Item-based CF. The recommendation accuracy reaches its global optimum when the number of neighbors K≥60 and the recommendation list size N=20. Ablation studies further confirm that integrating all three implicit feedback dimensions yields the best performance. This study provides an effective, scalable, and intelligent solution for university library management.